Weibull Regression with LSTM


I am trying to perform Regression with target distribution assumed to be Weibull.

The input is a time-ordered sequence with each timestamp having a regression target.

At each timestamp, the input is a mix of categorical and numerical values.

I transformed categorical variables using embedding and numerical values using an FFN.

When I train the network using LSTM, I am getting constant values as output, irrespective of the input.

Not sure how to debug/inspect this problem.

I am new to Pytorch, so I am confused.

Any suggestions wil be helpful.